target-i386/ops_sse.h

   1 /*
   2  *  MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI support
   3  *
   4  *  Copyright (c) 2005 Fabrice Bellard
   5  *  Copyright (c) 2008 Intel Corporation  <andrew.zaborowski@intel.com>
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  19  */
  20 #if SHIFT == 0
  21 #define Reg MMXReg
  22 #define XMM_ONLY(...)
  23 #define B(n) MMX_B(n)
  24 #define W(n) MMX_W(n)
  25 #define L(n) MMX_L(n)
  26 #define Q(n) q
  27 #define SUFFIX _mmx
  28 #else
  29 #define Reg XMMReg
  30 #define XMM_ONLY(...) __VA_ARGS__
  31 #define B(n) XMM_B(n)
  32 #define W(n) XMM_W(n)
  33 #define L(n) XMM_L(n)
  34 #define Q(n) XMM_Q(n)
  35 #define SUFFIX _xmm
  36 #endif
  37
  38 void glue(helper_psrlw, SUFFIX)(Reg *d, Reg *s)
  39 {
  40     int shift;
  41
  42     if (s->Q(0) > 15) {
  43         d->Q(0) = 0;
  44 #if SHIFT == 1
  45         d->Q(1) = 0;
  46 #endif
  47     } else {
  48         shift = s->B(0);
  49         d->W(0) >>= shift;
  50         d->W(1) >>= shift;
  51         d->W(2) >>= shift;
  52         d->W(3) >>= shift;
  53 #if SHIFT == 1
  54         d->W(4) >>= shift;
  55         d->W(5) >>= shift;
  56         d->W(6) >>= shift;
  57         d->W(7) >>= shift;
  58 #endif
  59     }
  60 }
  61
  62 void glue(helper_psraw, SUFFIX)(Reg *d, Reg *s)
  63 {
  64     int shift;
  65
  66     if (s->Q(0) > 15) {
  67         shift = 15;
  68     } else {
  69         shift = s->B(0);
  70     }
  71     d->W(0) = (int16_t)d->W(0) >> shift;
  72     d->W(1) = (int16_t)d->W(1) >> shift;
  73     d->W(2) = (int16_t)d->W(2) >> shift;
  74     d->W(3) = (int16_t)d->W(3) >> shift;
  75 #if SHIFT == 1
  76     d->W(4) = (int16_t)d->W(4) >> shift;
  77     d->W(5) = (int16_t)d->W(5) >> shift;
  78     d->W(6) = (int16_t)d->W(6) >> shift;
  79     d->W(7) = (int16_t)d->W(7) >> shift;
  80 #endif
  81 }
  82
  83 void glue(helper_psllw, SUFFIX)(Reg *d, Reg *s)
  84 {
  85     int shift;
  86
  87     if (s->Q(0) > 15) {
  88         d->Q(0) = 0;
  89 #if SHIFT == 1
  90         d->Q(1) = 0;
  91 #endif
  92     } else {
  93         shift = s->B(0);
  94         d->W(0) <<= shift;
  95         d->W(1) <<= shift;
  96         d->W(2) <<= shift;
  97         d->W(3) <<= shift;
  98 #if SHIFT == 1
  99         d->W(4) <<= shift;
 100         d->W(5) <<= shift;
 101         d->W(6) <<= shift;
 102         d->W(7) <<= shift;
 103 #endif
 104     }
 105 }
 106
 107 void glue(helper_psrld, SUFFIX)(Reg *d, Reg *s)
 108 {
 109     int shift;
 110
 111     if (s->Q(0) > 31) {
 112         d->Q(0) = 0;
 113 #if SHIFT == 1
 114         d->Q(1) = 0;
 115 #endif
 116     } else {
 117         shift = s->B(0);
 118         d->L(0) >>= shift;
 119         d->L(1) >>= shift;
 120 #if SHIFT == 1
 121         d->L(2) >>= shift;
 122         d->L(3) >>= shift;
 123 #endif
 124     }
 125 }
 126
 127 void glue(helper_psrad, SUFFIX)(Reg *d, Reg *s)
 128 {
 129     int shift;
 130
 131     if (s->Q(0) > 31) {
 132         shift = 31;
 133     } else {
 134         shift = s->B(0);
 135     }
 136     d->L(0) = (int32_t)d->L(0) >> shift;
 137     d->L(1) = (int32_t)d->L(1) >> shift;
 138 #if SHIFT == 1
 139     d->L(2) = (int32_t)d->L(2) >> shift;
 140     d->L(3) = (int32_t)d->L(3) >> shift;
 141 #endif
 142 }
 143
 144 void glue(helper_pslld, SUFFIX)(Reg *d, Reg *s)
 145 {
 146     int shift;
 147
 148     if (s->Q(0) > 31) {
 149         d->Q(0) = 0;
 150 #if SHIFT == 1
 151         d->Q(1) = 0;
 152 #endif
 153     } else {
 154         shift = s->B(0);
 155         d->L(0) <<= shift;
 156         d->L(1) <<= shift;
 157 #if SHIFT == 1
 158         d->L(2) <<= shift;
 159         d->L(3) <<= shift;
 160 #endif
 161     }
 162 }
 163
 164 void glue(helper_psrlq, SUFFIX)(Reg *d, Reg *s)
 165 {
 166     int shift;
 167
 168     if (s->Q(0) > 63) {
 169         d->Q(0) = 0;
 170 #if SHIFT == 1
 171         d->Q(1) = 0;
 172 #endif
 173     } else {
 174         shift = s->B(0);
 175         d->Q(0) >>= shift;
 176 #if SHIFT == 1
 177         d->Q(1) >>= shift;
 178 #endif
 179     }
 180 }
 181
 182 void glue(helper_psllq, SUFFIX)(Reg *d, Reg *s)
 183 {
 184     int shift;
 185
 186     if (s->Q(0) > 63) {
 187         d->Q(0) = 0;
 188 #if SHIFT == 1
 189         d->Q(1) = 0;
 190 #endif
 191     } else {
 192         shift = s->B(0);
 193         d->Q(0) <<= shift;
 194 #if SHIFT == 1
 195         d->Q(1) <<= shift;
 196 #endif
 197     }
 198 }
 199
 200 #if SHIFT == 1
 201 void glue(helper_psrldq, SUFFIX)(Reg *d, Reg *s)
 202 {
 203     int shift, i;
 204
 205     shift = s->L(0);
 206     if (shift > 16)
 207         shift = 16;
 208     for(i = 0; i < 16 - shift; i++)
 209         d->B(i) = d->B(i + shift);
 210     for(i = 16 - shift; i < 16; i++)
 211         d->B(i) = 0;
 212 }
 213
 214 void glue(helper_pslldq, SUFFIX)(Reg *d, Reg *s)
 215 {
 216     int shift, i;
 217
 218     shift = s->L(0);
 219     if (shift > 16)
 220         shift = 16;
 221     for(i = 15; i >= shift; i--)
 222         d->B(i) = d->B(i - shift);
 223     for(i = 0; i < shift; i++)
 224         d->B(i) = 0;
 225 }
 226 #endif
 227
 228 #define SSE_HELPER_B(name, F)\
 229 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 230 {\
 231     d->B(0) = F(d->B(0), s->B(0));\
 232     d->B(1) = F(d->B(1), s->B(1));\
 233     d->B(2) = F(d->B(2), s->B(2));\
 234     d->B(3) = F(d->B(3), s->B(3));\
 235     d->B(4) = F(d->B(4), s->B(4));\
 236     d->B(5) = F(d->B(5), s->B(5));\
 237     d->B(6) = F(d->B(6), s->B(6));\
 238     d->B(7) = F(d->B(7), s->B(7));\
 239     XMM_ONLY(\
 240     d->B(8) = F(d->B(8), s->B(8));\
 241     d->B(9) = F(d->B(9), s->B(9));\
 242     d->B(10) = F(d->B(10), s->B(10));\
 243     d->B(11) = F(d->B(11), s->B(11));\
 244     d->B(12) = F(d->B(12), s->B(12));\
 245     d->B(13) = F(d->B(13), s->B(13));\
 246     d->B(14) = F(d->B(14), s->B(14));\
 247     d->B(15) = F(d->B(15), s->B(15));\
 248     )\
 249 }
 250
 251 #define SSE_HELPER_W(name, F)\
 252 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 253 {\
 254     d->W(0) = F(d->W(0), s->W(0));\
 255     d->W(1) = F(d->W(1), s->W(1));\
 256     d->W(2) = F(d->W(2), s->W(2));\
 257     d->W(3) = F(d->W(3), s->W(3));\
 258     XMM_ONLY(\
 259     d->W(4) = F(d->W(4), s->W(4));\
 260     d->W(5) = F(d->W(5), s->W(5));\
 261     d->W(6) = F(d->W(6), s->W(6));\
 262     d->W(7) = F(d->W(7), s->W(7));\
 263     )\
 264 }
 265
 266 #define SSE_HELPER_L(name, F)\
 267 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 268 {\
 269     d->L(0) = F(d->L(0), s->L(0));\
 270     d->L(1) = F(d->L(1), s->L(1));\
 271     XMM_ONLY(\
 272     d->L(2) = F(d->L(2), s->L(2));\
 273     d->L(3) = F(d->L(3), s->L(3));\
 274     )\
 275 }
 276
 277 #define SSE_HELPER_Q(name, F)\
 278 void glue(name, SUFFIX) (Reg *d, Reg *s)\
 279 {\
 280     d->Q(0) = F(d->Q(0), s->Q(0));\
 281     XMM_ONLY(\
 282     d->Q(1) = F(d->Q(1), s->Q(1));\
 283     )\
 284 }
 285
 286 #if SHIFT == 0
 287 static inline int satub(int x)
 288 {
 289     if (x < 0)
 290         return 0;
 291     else if (x > 255)
 292         return 255;
 293     else
 294         return x;
 295 }
 296
 297 static inline int satuw(int x)
 298 {
 299     if (x < 0)
 300         return 0;
 301     else if (x > 65535)
 302         return 65535;
 303     else
 304         return x;
 305 }
 306
 307 static inline int satsb(int x)
 308 {
 309     if (x < -128)
 310         return -128;
 311     else if (x > 127)
 312         return 127;
 313     else
 314         return x;
 315 }
 316
 317 static inline int satsw(int x)
 318 {
 319     if (x < -32768)
 320         return -32768;
 321     else if (x > 32767)
 322         return 32767;
 323     else
 324         return x;
 325 }
 326
 327 #define FADD(a, b) ((a) + (b))
 328 #define FADDUB(a, b) satub((a) + (b))
 329 #define FADDUW(a, b) satuw((a) + (b))
 330 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
 331 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
 332
 333 #define FSUB(a, b) ((a) - (b))
 334 #define FSUBUB(a, b) satub((a) - (b))
 335 #define FSUBUW(a, b) satuw((a) - (b))
 336 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
 337 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
 338 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
 339 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
 340 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
 341 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
 342
 343 #define FAND(a, b) (a) & (b)
 344 #define FANDN(a, b) ((~(a)) & (b))
 345 #define FOR(a, b) (a) | (b)
 346 #define FXOR(a, b) (a) ^ (b)
 347
 348 #define FCMPGTB(a, b) (int8_t)(a) > (int8_t)(b) ? -1 : 0
 349 #define FCMPGTW(a, b) (int16_t)(a) > (int16_t)(b) ? -1 : 0
 350 #define FCMPGTL(a, b) (int32_t)(a) > (int32_t)(b) ? -1 : 0
 351 #define FCMPEQ(a, b) (a) == (b) ? -1 : 0
 352
 353 #define FMULLW(a, b) (a) * (b)
 354 #define FMULHRW(a, b) ((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16
 355 #define FMULHUW(a, b) (a) * (b) >> 16
 356 #define FMULHW(a, b) (int16_t)(a) * (int16_t)(b) >> 16
 357
 358 #define FAVG(a, b) ((a) + (b) + 1) >> 1
 359 #endif
 360
 361 SSE_HELPER_B(helper_paddb, FADD)
 362 SSE_HELPER_W(helper_paddw, FADD)
 363 SSE_HELPER_L(helper_paddl, FADD)
 364 SSE_HELPER_Q(helper_paddq, FADD)
 365
 366 SSE_HELPER_B(helper_psubb, FSUB)
 367 SSE_HELPER_W(helper_psubw, FSUB)
 368 SSE_HELPER_L(helper_psubl, FSUB)
 369 SSE_HELPER_Q(helper_psubq, FSUB)
 370
 371 SSE_HELPER_B(helper_paddusb, FADDUB)
 372 SSE_HELPER_B(helper_paddsb, FADDSB)
 373 SSE_HELPER_B(helper_psubusb, FSUBUB)
 374 SSE_HELPER_B(helper_psubsb, FSUBSB)
 375
 376 SSE_HELPER_W(helper_paddusw, FADDUW)
 377 SSE_HELPER_W(helper_paddsw, FADDSW)
 378 SSE_HELPER_W(helper_psubusw, FSUBUW)
 379 SSE_HELPER_W(helper_psubsw, FSUBSW)
 380
 381 SSE_HELPER_B(helper_pminub, FMINUB)
 382 SSE_HELPER_B(helper_pmaxub, FMAXUB)
 383
 384 SSE_HELPER_W(helper_pminsw, FMINSW)
 385 SSE_HELPER_W(helper_pmaxsw, FMAXSW)
 386
 387 SSE_HELPER_Q(helper_pand, FAND)
 388 SSE_HELPER_Q(helper_pandn, FANDN)
 389 SSE_HELPER_Q(helper_por, FOR)
 390 SSE_HELPER_Q(helper_pxor, FXOR)
 391
 392 SSE_HELPER_B(helper_pcmpgtb, FCMPGTB)
 393 SSE_HELPER_W(helper_pcmpgtw, FCMPGTW)
 394 SSE_HELPER_L(helper_pcmpgtl, FCMPGTL)
 395
 396 SSE_HELPER_B(helper_pcmpeqb, FCMPEQ)
 397 SSE_HELPER_W(helper_pcmpeqw, FCMPEQ)
 398 SSE_HELPER_L(helper_pcmpeql, FCMPEQ)
 399
 400 SSE_HELPER_W(helper_pmullw, FMULLW)
 401 #if SHIFT == 0
 402 SSE_HELPER_W(helper_pmulhrw, FMULHRW)
 403 #endif
 404 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
 405 SSE_HELPER_W(helper_pmulhw, FMULHW)
 406
 407 SSE_HELPER_B(helper_pavgb, FAVG)
 408 SSE_HELPER_W(helper_pavgw, FAVG)
 409
 410 void glue(helper_pmuludq, SUFFIX) (Reg *d, Reg *s)
 411 {
 412     d->Q(0) = (uint64_t)s->L(0) * (uint64_t)d->L(0);
 413 #if SHIFT == 1
 414     d->Q(1) = (uint64_t)s->L(2) * (uint64_t)d->L(2);
 415 #endif
 416 }
 417
 418 void glue(helper_pmaddwd, SUFFIX) (Reg *d, Reg *s)
 419 {
 420     int i;
 421
 422     for(i = 0; i < (2 << SHIFT); i++) {
 423         d->L(i) = (int16_t)s->W(2*i) * (int16_t)d->W(2*i) +
 424             (int16_t)s->W(2*i+1) * (int16_t)d->W(2*i+1);
 425     }
 426 }
 427
 428 #if SHIFT == 0
 429 static inline int abs1(int a)
 430 {
 431     if (a < 0)
 432         return -a;
 433     else
 434         return a;
 435 }
 436 #endif
 437 void glue(helper_psadbw, SUFFIX) (Reg *d, Reg *s)
 438 {
 439     unsigned int val;
 440
 441     val = 0;
 442     val += abs1(d->B(0) - s->B(0));
 443     val += abs1(d->B(1) - s->B(1));
 444     val += abs1(d->B(2) - s->B(2));
 445     val += abs1(d->B(3) - s->B(3));
 446     val += abs1(d->B(4) - s->B(4));
 447     val += abs1(d->B(5) - s->B(5));
 448     val += abs1(d->B(6) - s->B(6));
 449     val += abs1(d->B(7) - s->B(7));
 450     d->Q(0) = val;
 451 #if SHIFT == 1
 452     val = 0;
 453     val += abs1(d->B(8) - s->B(8));
 454     val += abs1(d->B(9) - s->B(9));
 455     val += abs1(d->B(10) - s->B(10));
 456     val += abs1(d->B(11) - s->B(11));
 457     val += abs1(d->B(12) - s->B(12));
 458     val += abs1(d->B(13) - s->B(13));
 459     val += abs1(d->B(14) - s->B(14));
 460     val += abs1(d->B(15) - s->B(15));
 461     d->Q(1) = val;
 462 #endif
 463 }
 464
 465 void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s, target_ulong a0)
 466 {
 467     int i;
 468     for(i = 0; i < (8 << SHIFT); i++) {
 469         if (s->B(i) & 0x80)
 470             stb(a0 + i, d->B(i));
 471     }
 472 }
 473
 474 void glue(helper_movl_mm_T0, SUFFIX) (Reg *d, uint32_t val)
 475 {
 476     d->L(0) = val;
 477     d->L(1) = 0;
 478 #if SHIFT == 1
 479     d->Q(1) = 0;
 480 #endif
 481 }
 482
 483 #ifdef TARGET_X86_64
 484 void glue(helper_movq_mm_T0, SUFFIX) (Reg *d, uint64_t val)
 485 {
 486     d->Q(0) = val;
 487 #if SHIFT == 1
 488     d->Q(1) = 0;
 489 #endif
 490 }
 491 #endif
 492
 493 #if SHIFT == 0
 494 void glue(helper_pshufw, SUFFIX) (Reg *d, Reg *s, int order)
 495 {
 496     Reg r;
 497     r.W(0) = s->W(order & 3);
 498     r.W(1) = s->W((order >> 2) & 3);
 499     r.W(2) = s->W((order >> 4) & 3);
 500     r.W(3) = s->W((order >> 6) & 3);
 501     *d = r;
 502 }
 503 #else
 504 void helper_shufps(Reg *d, Reg *s, int order)
 505 {
 506     Reg r;
 507     r.L(0) = d->L(order & 3);
 508     r.L(1) = d->L((order >> 2) & 3);
 509     r.L(2) = s->L((order >> 4) & 3);
 510     r.L(3) = s->L((order >> 6) & 3);
 511     *d = r;
 512 }
 513
 514 void helper_shufpd(Reg *d, Reg *s, int order)
 515 {
 516     Reg r;
 517     r.Q(0) = d->Q(order & 1);
 518     r.Q(1) = s->Q((order >> 1) & 1);
 519     *d = r;
 520 }
 521
 522 void glue(helper_pshufd, SUFFIX) (Reg *d, Reg *s, int order)
 523 {
 524     Reg r;
 525     r.L(0) = s->L(order & 3);
 526     r.L(1) = s->L((order >> 2) & 3);
 527     r.L(2) = s->L((order >> 4) & 3);
 528     r.L(3) = s->L((order >> 6) & 3);
 529     *d = r;
 530 }
 531
 532 void glue(helper_pshuflw, SUFFIX) (Reg *d, Reg *s, int order)
 533 {
 534     Reg r;
 535     r.W(0) = s->W(order & 3);
 536     r.W(1) = s->W((order >> 2) & 3);
 537     r.W(2) = s->W((order >> 4) & 3);
 538     r.W(3) = s->W((order >> 6) & 3);
 539     r.Q(1) = s->Q(1);
 540     *d = r;
 541 }
 542
 543 void glue(helper_pshufhw, SUFFIX) (Reg *d, Reg *s, int order)
 544 {
 545     Reg r;
 546     r.Q(0) = s->Q(0);
 547     r.W(4) = s->W(4 + (order & 3));
 548     r.W(5) = s->W(4 + ((order >> 2) & 3));
 549     r.W(6) = s->W(4 + ((order >> 4) & 3));
 550     r.W(7) = s->W(4 + ((order >> 6) & 3));
 551     *d = r;
 552 }
 553 #endif
 554
 555 #if SHIFT == 1
 556 /* FPU ops */
 557 /* XXX: not accurate */
 558
 559 #define SSE_HELPER_S(name, F)\
 560 void helper_ ## name ## ps (Reg *d, Reg *s)\
 561 {\
 562     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 563     d->XMM_S(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 564     d->XMM_S(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 565     d->XMM_S(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 566 }\
 567 \
 568 void helper_ ## name ## ss (Reg *d, Reg *s)\
 569 {\
 570     d->XMM_S(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 571 }\
 572 void helper_ ## name ## pd (Reg *d, Reg *s)\
 573 {\
 574     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 575     d->XMM_D(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 576 }\
 577 \
 578 void helper_ ## name ## sd (Reg *d, Reg *s)\
 579 {\
 580     d->XMM_D(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 581 }
 582
 583 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
 584 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
 585 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
 586 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
 587 #define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
 588 #define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
 589 #define FPU_SQRT(size, a, b) float ## size ## _sqrt(b, &env->sse_status)
 590
 591 SSE_HELPER_S(add, FPU_ADD)
 592 SSE_HELPER_S(sub, FPU_SUB)
 593 SSE_HELPER_S(mul, FPU_MUL)
 594 SSE_HELPER_S(div, FPU_DIV)
 595 SSE_HELPER_S(min, FPU_MIN)
 596 SSE_HELPER_S(max, FPU_MAX)
 597 SSE_HELPER_S(sqrt, FPU_SQRT)
 598
 599
 600 /* float to float conversions */
 601 void helper_cvtps2pd(Reg *d, Reg *s)
 602 {
 603     float32 s0, s1;
 604     s0 = s->XMM_S(0);
 605     s1 = s->XMM_S(1);
 606     d->XMM_D(0) = float32_to_float64(s0, &env->sse_status);
 607     d->XMM_D(1) = float32_to_float64(s1, &env->sse_status);
 608 }
 609
 610 void helper_cvtpd2ps(Reg *d, Reg *s)
 611 {
 612     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 613     d->XMM_S(1) = float64_to_float32(s->XMM_D(1), &env->sse_status);
 614     d->Q(1) = 0;
 615 }
 616
 617 void helper_cvtss2sd(Reg *d, Reg *s)
 618 {
 619     d->XMM_D(0) = float32_to_float64(s->XMM_S(0), &env->sse_status);
 620 }
 621
 622 void helper_cvtsd2ss(Reg *d, Reg *s)
 623 {
 624     d->XMM_S(0) = float64_to_float32(s->XMM_D(0), &env->sse_status);
 625 }
 626
 627 /* integer to float */
 628 void helper_cvtdq2ps(Reg *d, Reg *s)
 629 {
 630     d->XMM_S(0) = int32_to_float32(s->XMM_L(0), &env->sse_status);
 631     d->XMM_S(1) = int32_to_float32(s->XMM_L(1), &env->sse_status);
 632     d->XMM_S(2) = int32_to_float32(s->XMM_L(2), &env->sse_status);
 633     d->XMM_S(3) = int32_to_float32(s->XMM_L(3), &env->sse_status);
 634 }
 635
 636 void helper_cvtdq2pd(Reg *d, Reg *s)
 637 {
 638     int32_t l0, l1;
 639     l0 = (int32_t)s->XMM_L(0);
 640     l1 = (int32_t)s->XMM_L(1);
 641     d->XMM_D(0) = int32_to_float64(l0, &env->sse_status);
 642     d->XMM_D(1) = int32_to_float64(l1, &env->sse_status);
 643 }
 644
 645 void helper_cvtpi2ps(XMMReg *d, MMXReg *s)
 646 {
 647     d->XMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
 648     d->XMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
 649 }
 650
 651 void helper_cvtpi2pd(XMMReg *d, MMXReg *s)
 652 {
 653     d->XMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
 654     d->XMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
 655 }
 656
 657 void helper_cvtsi2ss(XMMReg *d, uint32_t val)
 658 {
 659     d->XMM_S(0) = int32_to_float32(val, &env->sse_status);
 660 }
 661
 662 void helper_cvtsi2sd(XMMReg *d, uint32_t val)
 663 {
 664     d->XMM_D(0) = int32_to_float64(val, &env->sse_status);
 665 }
 666
 667 #ifdef TARGET_X86_64
 668 void helper_cvtsq2ss(XMMReg *d, uint64_t val)
 669 {
 670     d->XMM_S(0) = int64_to_float32(val, &env->sse_status);
 671 }
 672
 673 void helper_cvtsq2sd(XMMReg *d, uint64_t val)
 674 {
 675     d->XMM_D(0) = int64_to_float64(val, &env->sse_status);
 676 }
 677 #endif
 678
 679 /* float to integer */
 680 void helper_cvtps2dq(XMMReg *d, XMMReg *s)
 681 {
 682     d->XMM_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 683     d->XMM_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 684     d->XMM_L(2) = float32_to_int32(s->XMM_S(2), &env->sse_status);
 685     d->XMM_L(3) = float32_to_int32(s->XMM_S(3), &env->sse_status);
 686 }
 687
 688 void helper_cvtpd2dq(XMMReg *d, XMMReg *s)
 689 {
 690     d->XMM_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 691     d->XMM_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 692     d->XMM_Q(1) = 0;
 693 }
 694
 695 void helper_cvtps2pi(MMXReg *d, XMMReg *s)
 696 {
 697     d->MMX_L(0) = float32_to_int32(s->XMM_S(0), &env->sse_status);
 698     d->MMX_L(1) = float32_to_int32(s->XMM_S(1), &env->sse_status);
 699 }
 700
 701 void helper_cvtpd2pi(MMXReg *d, XMMReg *s)
 702 {
 703     d->MMX_L(0) = float64_to_int32(s->XMM_D(0), &env->sse_status);
 704     d->MMX_L(1) = float64_to_int32(s->XMM_D(1), &env->sse_status);
 705 }
 706
 707 int32_t helper_cvtss2si(XMMReg *s)
 708 {
 709     return float32_to_int32(s->XMM_S(0), &env->sse_status);
 710 }
 711
 712 int32_t helper_cvtsd2si(XMMReg *s)
 713 {
 714     return float64_to_int32(s->XMM_D(0), &env->sse_status);
 715 }
 716
 717 #ifdef TARGET_X86_64
 718 int64_t helper_cvtss2sq(XMMReg *s)
 719 {
 720     return float32_to_int64(s->XMM_S(0), &env->sse_status);
 721 }
 722
 723 int64_t helper_cvtsd2sq(XMMReg *s)
 724 {
 725     return float64_to_int64(s->XMM_D(0), &env->sse_status);
 726 }
 727 #endif
 728
 729 /* float to integer truncated */
 730 void helper_cvttps2dq(XMMReg *d, XMMReg *s)
 731 {
 732     d->XMM_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 733     d->XMM_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 734     d->XMM_L(2) = float32_to_int32_round_to_zero(s->XMM_S(2), &env->sse_status);
 735     d->XMM_L(3) = float32_to_int32_round_to_zero(s->XMM_S(3), &env->sse_status);
 736 }
 737
 738 void helper_cvttpd2dq(XMMReg *d, XMMReg *s)
 739 {
 740     d->XMM_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 741     d->XMM_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 742     d->XMM_Q(1) = 0;
 743 }
 744
 745 void helper_cvttps2pi(MMXReg *d, XMMReg *s)
 746 {
 747     d->MMX_L(0) = float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 748     d->MMX_L(1) = float32_to_int32_round_to_zero(s->XMM_S(1), &env->sse_status);
 749 }
 750
 751 void helper_cvttpd2pi(MMXReg *d, XMMReg *s)
 752 {
 753     d->MMX_L(0) = float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 754     d->MMX_L(1) = float64_to_int32_round_to_zero(s->XMM_D(1), &env->sse_status);
 755 }
 756
 757 int32_t helper_cvttss2si(XMMReg *s)
 758 {
 759     return float32_to_int32_round_to_zero(s->XMM_S(0), &env->sse_status);
 760 }
 761
 762 int32_t helper_cvttsd2si(XMMReg *s)
 763 {
 764     return float64_to_int32_round_to_zero(s->XMM_D(0), &env->sse_status);
 765 }
 766
 767 #ifdef TARGET_X86_64
 768 int64_t helper_cvttss2sq(XMMReg *s)
 769 {
 770     return float32_to_int64_round_to_zero(s->XMM_S(0), &env->sse_status);
 771 }
 772
 773 int64_t helper_cvttsd2sq(XMMReg *s)
 774 {
 775     return float64_to_int64_round_to_zero(s->XMM_D(0), &env->sse_status);
 776 }
 777 #endif
 778
 779 void helper_rsqrtps(XMMReg *d, XMMReg *s)
 780 {
 781     d->XMM_S(0) = float32_div(float32_one,
 782                               float32_sqrt(s->XMM_S(0), &env->sse_status),
 783                               &env->sse_status);
 784     d->XMM_S(1) = float32_div(float32_one,
 785                               float32_sqrt(s->XMM_S(1), &env->sse_status),
 786                               &env->sse_status);
 787     d->XMM_S(2) = float32_div(float32_one,
 788                               float32_sqrt(s->XMM_S(2), &env->sse_status),
 789                               &env->sse_status);
 790     d->XMM_S(3) = float32_div(float32_one,
 791                               float32_sqrt(s->XMM_S(3), &env->sse_status),
 792                               &env->sse_status);
 793 }
 794
 795 void helper_rsqrtss(XMMReg *d, XMMReg *s)
 796 {
 797     d->XMM_S(0) = float32_div(float32_one,
 798                               float32_sqrt(s->XMM_S(0), &env->sse_status),
 799                               &env->sse_status);
 800 }
 801
 802 void helper_rcpps(XMMReg *d, XMMReg *s)
 803 {
 804     d->XMM_S(0) = float32_div(float32_one, s->XMM_S(0), &env->sse_status);
 805     d->XMM_S(1) = float32_div(float32_one, s->XMM_S(1), &env->sse_status);
 806     d->XMM_S(2) = float32_div(float32_one, s->XMM_S(2), &env->sse_status);
 807     d->XMM_S(3) = float32_div(float32_one, s->XMM_S(3), &env->sse_status);
 808 }
 809
 810 void helper_rcpss(XMMReg *d, XMMReg *s)
 811 {
 812     d->XMM_S(0) = float32_div(float32_one, s->XMM_S(0), &env->sse_status);
 813 }
 814
 815 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
 816 {
 817     uint64_t mask;
 818
 819     if (len == 0) {
 820         mask = ~0LL;
 821     } else {
 822         mask = (1ULL << len) - 1;
 823     }
 824     return (src >> shift) & mask;
 825 }
 826
 827 void helper_extrq_r(XMMReg *d, XMMReg *s)
 828 {
 829     d->XMM_Q(0) = helper_extrq(d->XMM_Q(0), s->XMM_B(1), s->XMM_B(0));
 830 }
 831
 832 void helper_extrq_i(XMMReg *d, int index, int length)
 833 {
 834     d->XMM_Q(0) = helper_extrq(d->XMM_Q(0), index, length);
 835 }
 836
 837 static inline uint64_t helper_insertq(uint64_t src, int shift, int len)
 838 {
 839     uint64_t mask;
 840
 841     if (len == 0) {
 842         mask = ~0ULL;
 843     } else {
 844         mask = (1ULL << len) - 1;
 845     }
 846     return (src & ~(mask << shift)) | ((src & mask) << shift);
 847 }
 848
 849 void helper_insertq_r(XMMReg *d, XMMReg *s)
 850 {
 851     d->XMM_Q(0) = helper_insertq(s->XMM_Q(0), s->XMM_B(9), s->XMM_B(8));
 852 }
 853
 854 void helper_insertq_i(XMMReg *d, int index, int length)
 855 {
 856     d->XMM_Q(0) = helper_insertq(d->XMM_Q(0), index, length);
 857 }
 858
 859 void helper_haddps(XMMReg *d, XMMReg *s)
 860 {
 861     XMMReg r;
 862     r.XMM_S(0) = float32_add(d->XMM_S(0), d->XMM_S(1), &env->sse_status);
 863     r.XMM_S(1) = float32_add(d->XMM_S(2), d->XMM_S(3), &env->sse_status);
 864     r.XMM_S(2) = float32_add(s->XMM_S(0), s->XMM_S(1), &env->sse_status);
 865     r.XMM_S(3) = float32_add(s->XMM_S(2), s->XMM_S(3), &env->sse_status);
 866     *d = r;
 867 }
 868
 869 void helper_haddpd(XMMReg *d, XMMReg *s)
 870 {
 871     XMMReg r;
 872     r.XMM_D(0) = float64_add(d->XMM_D(0), d->XMM_D(1), &env->sse_status);
 873     r.XMM_D(1) = float64_add(s->XMM_D(0), s->XMM_D(1), &env->sse_status);
 874     *d = r;
 875 }
 876
 877 void helper_hsubps(XMMReg *d, XMMReg *s)
 878 {
 879     XMMReg r;
 880     r.XMM_S(0) = float32_sub(d->XMM_S(0), d->XMM_S(1), &env->sse_status);
 881     r.XMM_S(1) = float32_sub(d->XMM_S(2), d->XMM_S(3), &env->sse_status);
 882     r.XMM_S(2) = float32_sub(s->XMM_S(0), s->XMM_S(1), &env->sse_status);
 883     r.XMM_S(3) = float32_sub(s->XMM_S(2), s->XMM_S(3), &env->sse_status);
 884     *d = r;
 885 }
 886
 887 void helper_hsubpd(XMMReg *d, XMMReg *s)
 888 {
 889     XMMReg r;
 890     r.XMM_D(0) = float64_sub(d->XMM_D(0), d->XMM_D(1), &env->sse_status);
 891     r.XMM_D(1) = float64_sub(s->XMM_D(0), s->XMM_D(1), &env->sse_status);
 892     *d = r;
 893 }
 894
 895 void helper_addsubps(XMMReg *d, XMMReg *s)
 896 {
 897     d->XMM_S(0) = float32_sub(d->XMM_S(0), s->XMM_S(0), &env->sse_status);
 898     d->XMM_S(1) = float32_add(d->XMM_S(1), s->XMM_S(1), &env->sse_status);
 899     d->XMM_S(2) = float32_sub(d->XMM_S(2), s->XMM_S(2), &env->sse_status);
 900     d->XMM_S(3) = float32_add(d->XMM_S(3), s->XMM_S(3), &env->sse_status);
 901 }
 902
 903 void helper_addsubpd(XMMReg *d, XMMReg *s)
 904 {
 905     d->XMM_D(0) = float64_sub(d->XMM_D(0), s->XMM_D(0), &env->sse_status);
 906     d->XMM_D(1) = float64_add(d->XMM_D(1), s->XMM_D(1), &env->sse_status);
 907 }
 908
 909 /* XXX: unordered */
 910 #define SSE_HELPER_CMP(name, F)\
 911 void helper_ ## name ## ps (Reg *d, Reg *s)\
 912 {\
 913     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 914     d->XMM_L(1) = F(32, d->XMM_S(1), s->XMM_S(1));\
 915     d->XMM_L(2) = F(32, d->XMM_S(2), s->XMM_S(2));\
 916     d->XMM_L(3) = F(32, d->XMM_S(3), s->XMM_S(3));\
 917 }\
 918 \
 919 void helper_ ## name ## ss (Reg *d, Reg *s)\
 920 {\
 921     d->XMM_L(0) = F(32, d->XMM_S(0), s->XMM_S(0));\
 922 }\
 923 void helper_ ## name ## pd (Reg *d, Reg *s)\
 924 {\
 925     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 926     d->XMM_Q(1) = F(64, d->XMM_D(1), s->XMM_D(1));\
 927 }\
 928 \
 929 void helper_ ## name ## sd (Reg *d, Reg *s)\
 930 {\
 931     d->XMM_Q(0) = F(64, d->XMM_D(0), s->XMM_D(0));\
 932 }
 933
 934 #define FPU_CMPEQ(size, a, b) float ## size ## _eq_quiet(a, b, &env->sse_status) ? -1 : 0
 935 #define FPU_CMPLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? -1 : 0
 936 #define FPU_CMPLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? -1 : 0
 937 #define FPU_CMPUNORD(size, a, b) float ## size ## _unordered_quiet(a, b, &env->sse_status) ? - 1 : 0
 938 #define FPU_CMPNEQ(size, a, b) float ## size ## _eq_quiet(a, b, &env->sse_status) ? 0 : -1
 939 #define FPU_CMPNLT(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? 0 : -1
 940 #define FPU_CMPNLE(size, a, b) float ## size ## _le(a, b, &env->sse_status) ? 0 : -1
 941 #define FPU_CMPORD(size, a, b) float ## size ## _unordered_quiet(a, b, &env->sse_status) ? 0 : -1
 942
 943 SSE_HELPER_CMP(cmpeq, FPU_CMPEQ)
 944 SSE_HELPER_CMP(cmplt, FPU_CMPLT)
 945 SSE_HELPER_CMP(cmple, FPU_CMPLE)
 946 SSE_HELPER_CMP(cmpunord, FPU_CMPUNORD)
 947 SSE_HELPER_CMP(cmpneq, FPU_CMPNEQ)
 948 SSE_HELPER_CMP(cmpnlt, FPU_CMPNLT)
 949 SSE_HELPER_CMP(cmpnle, FPU_CMPNLE)
 950 SSE_HELPER_CMP(cmpord, FPU_CMPORD)
 951
 952 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
 953
 954 void helper_ucomiss(Reg *d, Reg *s)
 955 {
 956     int ret;
 957     float32 s0, s1;
 958
 959     s0 = d->XMM_S(0);
 960     s1 = s->XMM_S(0);
 961     ret = float32_compare_quiet(s0, s1, &env->sse_status);
 962     CC_SRC = comis_eflags[ret + 1];
 963 }
 964
 965 void helper_comiss(Reg *d, Reg *s)
 966 {
 967     int ret;
 968     float32 s0, s1;
 969
 970     s0 = d->XMM_S(0);
 971     s1 = s->XMM_S(0);
 972     ret = float32_compare(s0, s1, &env->sse_status);
 973     CC_SRC = comis_eflags[ret + 1];
 974 }
 975
 976 void helper_ucomisd(Reg *d, Reg *s)
 977 {
 978     int ret;
 979     float64 d0, d1;
 980
 981     d0 = d->XMM_D(0);
 982     d1 = s->XMM_D(0);
 983     ret = float64_compare_quiet(d0, d1, &env->sse_status);
 984     CC_SRC = comis_eflags[ret + 1];
 985 }
 986
 987 void helper_comisd(Reg *d, Reg *s)
 988 {
 989     int ret;
 990     float64 d0, d1;
 991
 992     d0 = d->XMM_D(0);
 993     d1 = s->XMM_D(0);
 994     ret = float64_compare(d0, d1, &env->sse_status);
 995     CC_SRC = comis_eflags[ret + 1];
 996 }
 997
 998 uint32_t helper_movmskps(Reg *s)
 999 {
1000     int b0, b1, b2, b3;
1001     b0 = s->XMM_L(0) >> 31;
1002     b1 = s->XMM_L(1) >> 31;
1003     b2 = s->XMM_L(2) >> 31;
1004     b3 = s->XMM_L(3) >> 31;
1005     return b0 | (b1 << 1) | (b2 << 2) | (b3 << 3);
1006 }
1007
1008 uint32_t helper_movmskpd(Reg *s)
1009 {
1010     int b0, b1;
1011     b0 = s->XMM_L(1) >> 31;
1012     b1 = s->XMM_L(3) >> 31;
1013     return b0 | (b1 << 1);
1014 }
1015
1016 #endif
1017
1018 uint32_t glue(helper_pmovmskb, SUFFIX)(Reg *s)
1019 {
1020     uint32_t val;
1021     val = 0;
1022     val |= (s->B(0) >> 7);
1023     val |= (s->B(1) >> 6) & 0x02;
1024     val |= (s->B(2) >> 5) & 0x04;
1025     val |= (s->B(3) >> 4) & 0x08;
1026     val |= (s->B(4) >> 3) & 0x10;
1027     val |= (s->B(5) >> 2) & 0x20;
1028     val |= (s->B(6) >> 1) & 0x40;
1029     val |= (s->B(7)) & 0x80;
1030 #if SHIFT == 1
1031     val |= (s->B(8) << 1) & 0x0100;
1032     val |= (s->B(9) << 2) & 0x0200;
1033     val |= (s->B(10) << 3) & 0x0400;
1034     val |= (s->B(11) << 4) & 0x0800;
1035     val |= (s->B(12) << 5) & 0x1000;
1036     val |= (s->B(13) << 6) & 0x2000;
1037     val |= (s->B(14) << 7) & 0x4000;
1038     val |= (s->B(15) << 8) & 0x8000;
1039 #endif
1040     return val;
1041 }
1042
1043 void glue(helper_packsswb, SUFFIX) (Reg *d, Reg *s)
1044 {
1045     Reg r;
1046
1047     r.B(0) = satsb((int16_t)d->W(0));
1048     r.B(1) = satsb((int16_t)d->W(1));
1049     r.B(2) = satsb((int16_t)d->W(2));
1050     r.B(3) = satsb((int16_t)d->W(3));
1051 #if SHIFT == 1
1052     r.B(4) = satsb((int16_t)d->W(4));
1053     r.B(5) = satsb((int16_t)d->W(5));
1054     r.B(6) = satsb((int16_t)d->W(6));
1055     r.B(7) = satsb((int16_t)d->W(7));
1056 #endif
1057     r.B((4 << SHIFT) + 0) = satsb((int16_t)s->W(0));
1058     r.B((4 << SHIFT) + 1) = satsb((int16_t)s->W(1));
1059     r.B((4 << SHIFT) + 2) = satsb((int16_t)s->W(2));
1060     r.B((4 << SHIFT) + 3) = satsb((int16_t)s->W(3));
1061 #if SHIFT == 1
1062     r.B(12) = satsb((int16_t)s->W(4));
1063     r.B(13) = satsb((int16_t)s->W(5));
1064     r.B(14) = satsb((int16_t)s->W(6));
1065     r.B(15) = satsb((int16_t)s->W(7));
1066 #endif
1067     *d = r;
1068 }
1069
1070 void glue(helper_packuswb, SUFFIX) (Reg *d, Reg *s)
1071 {
1072     Reg r;
1073
1074     r.B(0) = satub((int16_t)d->W(0));
1075     r.B(1) = satub((int16_t)d->W(1));
1076     r.B(2) = satub((int16_t)d->W(2));
1077     r.B(3) = satub((int16_t)d->W(3));
1078 #if SHIFT == 1
1079     r.B(4) = satub((int16_t)d->W(4));
1080     r.B(5) = satub((int16_t)d->W(5));
1081     r.B(6) = satub((int16_t)d->W(6));
1082     r.B(7) = satub((int16_t)d->W(7));
1083 #endif
1084     r.B((4 << SHIFT) + 0) = satub((int16_t)s->W(0));
1085     r.B((4 << SHIFT) + 1) = satub((int16_t)s->W(1));
1086     r.B((4 << SHIFT) + 2) = satub((int16_t)s->W(2));
1087     r.B((4 << SHIFT) + 3) = satub((int16_t)s->W(3));
1088 #if SHIFT == 1
1089     r.B(12) = satub((int16_t)s->W(4));
1090     r.B(13) = satub((int16_t)s->W(5));
1091     r.B(14) = satub((int16_t)s->W(6));
1092     r.B(15) = satub((int16_t)s->W(7));
1093 #endif
1094     *d = r;
1095 }
1096
1097 void glue(helper_packssdw, SUFFIX) (Reg *d, Reg *s)
1098 {
1099     Reg r;
1100
1101     r.W(0) = satsw(d->L(0));
1102     r.W(1) = satsw(d->L(1));
1103 #if SHIFT == 1
1104     r.W(2) = satsw(d->L(2));
1105     r.W(3) = satsw(d->L(3));
1106 #endif
1107     r.W((2 << SHIFT) + 0) = satsw(s->L(0));
1108     r.W((2 << SHIFT) + 1) = satsw(s->L(1));
1109 #if SHIFT == 1
1110     r.W(6) = satsw(s->L(2));
1111     r.W(7) = satsw(s->L(3));
1112 #endif
1113     *d = r;
1114 }
1115
1116 #define UNPCK_OP(base_name, base)                               \
1117                                                                 \
1118 void glue(helper_punpck ## base_name ## bw, SUFFIX) (Reg *d, Reg *s)   \
1119 {                                                               \
1120     Reg r;                                              \
1121                                                                 \
1122     r.B(0) = d->B((base << (SHIFT + 2)) + 0);                   \
1123     r.B(1) = s->B((base << (SHIFT + 2)) + 0);                   \
1124     r.B(2) = d->B((base << (SHIFT + 2)) + 1);                   \
1125     r.B(3) = s->B((base << (SHIFT + 2)) + 1);                   \
1126     r.B(4) = d->B((base << (SHIFT + 2)) + 2);                   \
1127     r.B(5) = s->B((base << (SHIFT + 2)) + 2);                   \
1128     r.B(6) = d->B((base << (SHIFT + 2)) + 3);                   \
1129     r.B(7) = s->B((base << (SHIFT + 2)) + 3);                   \
1130 XMM_ONLY(                                                       \
1131     r.B(8) = d->B((base << (SHIFT + 2)) + 4);                   \
1132     r.B(9) = s->B((base << (SHIFT + 2)) + 4);                   \
1133     r.B(10) = d->B((base << (SHIFT + 2)) + 5);                  \
1134     r.B(11) = s->B((base << (SHIFT + 2)) + 5);                  \
1135     r.B(12) = d->B((base << (SHIFT + 2)) + 6);                  \
1136     r.B(13) = s->B((base << (SHIFT + 2)) + 6);                  \
1137     r.B(14) = d->B((base << (SHIFT + 2)) + 7);                  \
1138     r.B(15) = s->B((base << (SHIFT + 2)) + 7);                  \
1139 )                                                               \
1140     *d = r;                                                     \
1141 }                                                               \
1142                                                                 \
1143 void glue(helper_punpck ## base_name ## wd, SUFFIX) (Reg *d, Reg *s)   \
1144 {                                                               \
1145     Reg r;                                              \
1146                                                                 \
1147     r.W(0) = d->W((base << (SHIFT + 1)) + 0);                   \
1148     r.W(1) = s->W((base << (SHIFT + 1)) + 0);                   \
1149     r.W(2) = d->W((base << (SHIFT + 1)) + 1);                   \
1150     r.W(3) = s->W((base << (SHIFT + 1)) + 1);                   \
1151 XMM_ONLY(                                                       \
1152     r.W(4) = d->W((base << (SHIFT + 1)) + 2);                   \
1153     r.W(5) = s->W((base << (SHIFT + 1)) + 2);                   \
1154     r.W(6) = d->W((base << (SHIFT + 1)) + 3);                   \
1155     r.W(7) = s->W((base << (SHIFT + 1)) + 3);                   \
1156 )                                                               \
1157     *d = r;                                                     \
1158 }                                                               \
1159                                                                 \
1160 void glue(helper_punpck ## base_name ## dq, SUFFIX) (Reg *d, Reg *s)   \
1161 {                                                               \
1162     Reg r;                                              \
1163                                                                 \
1164     r.L(0) = d->L((base << SHIFT) + 0);                         \
1165     r.L(1) = s->L((base << SHIFT) + 0);                         \
1166 XMM_ONLY(                                                       \
1167     r.L(2) = d->L((base << SHIFT) + 1);                         \
1168     r.L(3) = s->L((base << SHIFT) + 1);                         \
1169 )                                                               \
1170     *d = r;                                                     \
1171 }                                                               \
1172                                                                 \
1173 XMM_ONLY(                                                       \
1174 void glue(helper_punpck ## base_name ## qdq, SUFFIX) (Reg *d, Reg *s)  \
1175 {                                                               \
1176     Reg r;                                              \
1177                                                                 \
1178     r.Q(0) = d->Q(base);                                        \
1179     r.Q(1) = s->Q(base);                                        \
1180     *d = r;                                                     \
1181 }                                                               \
1182 )
1183
1184 UNPCK_OP(l, 0)
1185 UNPCK_OP(h, 1)
1186
1187 /* 3DNow! float ops */
1188 #if SHIFT == 0
1189 void helper_pi2fd(MMXReg *d, MMXReg *s)
1190 {
1191     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1192     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1193 }
1194
1195 void helper_pi2fw(MMXReg *d, MMXReg *s)
1196 {
1197     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1198     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1199 }
1200
1201 void helper_pf2id(MMXReg *d, MMXReg *s)
1202 {
1203     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1204     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1205 }
1206
1207 void helper_pf2iw(MMXReg *d, MMXReg *s)
1208 {
1209     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status));
1210     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status));
1211 }
1212
1213 void helper_pfacc(MMXReg *d, MMXReg *s)
1214 {
1215     MMXReg r;
1216     r.MMX_S(0) = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1217     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1218     *d = r;
1219 }
1220
1221 void helper_pfadd(MMXReg *d, MMXReg *s)
1222 {
1223     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1224     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1225 }
1226
1227 void helper_pfcmpeq(MMXReg *d, MMXReg *s)
1228 {
1229     d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0), &env->mmx_status) ? -1 : 0;
1230     d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1), &env->mmx_status) ? -1 : 0;
1231 }
1232
1233 void helper_pfcmpge(MMXReg *d, MMXReg *s)
1234 {
1235     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1236     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1237 }
1238
1239 void helper_pfcmpgt(MMXReg *d, MMXReg *s)
1240 {
1241     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status) ? -1 : 0;
1242     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status) ? -1 : 0;
1243 }
1244
1245 void helper_pfmax(MMXReg *d, MMXReg *s)
1246 {
1247     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status))
1248         d->MMX_S(0) = s->MMX_S(0);
1249     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status))
1250         d->MMX_S(1) = s->MMX_S(1);
1251 }
1252
1253 void helper_pfmin(MMXReg *d, MMXReg *s)
1254 {
1255     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status))
1256         d->MMX_S(0) = s->MMX_S(0);
1257     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status))
1258         d->MMX_S(1) = s->MMX_S(1);
1259 }
1260
1261 void helper_pfmul(MMXReg *d, MMXReg *s)
1262 {
1263     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1264     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1265 }
1266
1267 void helper_pfnacc(MMXReg *d, MMXReg *s)
1268 {
1269     MMXReg r;
1270     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1271     r.MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1272     *d = r;
1273 }
1274
1275 void helper_pfpnacc(MMXReg *d, MMXReg *s)
1276 {
1277     MMXReg r;
1278     r.MMX_S(0) = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1279     r.MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1280     *d = r;
1281 }
1282
1283 void helper_pfrcp(MMXReg *d, MMXReg *s)
1284 {
1285     d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1286     d->MMX_S(1) = d->MMX_S(0);
1287 }
1288
1289 void helper_pfrsqrt(MMXReg *d, MMXReg *s)
1290 {
1291     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1292     d->MMX_S(1) = float32_div(float32_one,
1293                               float32_sqrt(d->MMX_S(1), &env->mmx_status),
1294                               &env->mmx_status);
1295     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1296     d->MMX_L(0) = d->MMX_L(1);
1297 }
1298
1299 void helper_pfsub(MMXReg *d, MMXReg *s)
1300 {
1301     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1302     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1303 }
1304
1305 void helper_pfsubr(MMXReg *d, MMXReg *s)
1306 {
1307     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1308     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1309 }
1310
1311 void helper_pswapd(MMXReg *d, MMXReg *s)
1312 {
1313     MMXReg r;
1314     r.MMX_L(0) = s->MMX_L(1);
1315     r.MMX_L(1) = s->MMX_L(0);
1316     *d = r;
1317 }
1318 #endif
1319
1320 /* SSSE3 op helpers */
1321 void glue(helper_pshufb, SUFFIX) (Reg *d, Reg *s)
1322 {
1323     int i;
1324     Reg r;
1325
1326     for (i = 0; i < (8 << SHIFT); i++)
1327         r.B(i) = (s->B(i) & 0x80) ? 0 : (d->B(s->B(i) & ((8 << SHIFT) - 1)));
1328
1329     *d = r;
1330 }
1331
1332 void glue(helper_phaddw, SUFFIX) (Reg *d, Reg *s)
1333 {
1334     d->W(0) = (int16_t)d->W(0) + (int16_t)d->W(1);
1335     d->W(1) = (int16_t)d->W(2) + (int16_t)d->W(3);
1336     XMM_ONLY(d->W(2) = (int16_t)d->W(4) + (int16_t)d->W(5));
1337     XMM_ONLY(d->W(3) = (int16_t)d->W(6) + (int16_t)d->W(7));
1338     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) + (int16_t)s->W(1);
1339     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) + (int16_t)s->W(3);
1340     XMM_ONLY(d->W(6) = (int16_t)s->W(4) + (int16_t)s->W(5));
1341     XMM_ONLY(d->W(7) = (int16_t)s->W(6) + (int16_t)s->W(7));
1342 }
1343
1344 void glue(helper_phaddd, SUFFIX) (Reg *d, Reg *s)
1345 {
1346     d->L(0) = (int32_t)d->L(0) + (int32_t)d->L(1);
1347     XMM_ONLY(d->L(1) = (int32_t)d->L(2) + (int32_t)d->L(3));
1348     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) + (int32_t)s->L(1);
1349     XMM_ONLY(d->L(3) = (int32_t)s->L(2) + (int32_t)s->L(3));
1350 }
1351
1352 void glue(helper_phaddsw, SUFFIX) (Reg *d, Reg *s)
1353 {
1354     d->W(0) = satsw((int16_t)d->W(0) + (int16_t)d->W(1));
1355     d->W(1) = satsw((int16_t)d->W(2) + (int16_t)d->W(3));
1356     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) + (int16_t)d->W(5)));
1357     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) + (int16_t)d->W(7)));
1358     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) + (int16_t)s->W(1));
1359     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) + (int16_t)s->W(3));
1360     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) + (int16_t)s->W(5)));
1361     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) + (int16_t)s->W(7)));
1362 }
1363
1364 void glue(helper_pmaddubsw, SUFFIX) (Reg *d, Reg *s)
1365 {
1366     d->W(0) = satsw((int8_t)s->B( 0) * (uint8_t)d->B( 0) +
1367                     (int8_t)s->B( 1) * (uint8_t)d->B( 1));
1368     d->W(1) = satsw((int8_t)s->B( 2) * (uint8_t)d->B( 2) +
1369                     (int8_t)s->B( 3) * (uint8_t)d->B( 3));
1370     d->W(2) = satsw((int8_t)s->B( 4) * (uint8_t)d->B( 4) +
1371                     (int8_t)s->B( 5) * (uint8_t)d->B( 5));
1372     d->W(3) = satsw((int8_t)s->B( 6) * (uint8_t)d->B( 6) +
1373                     (int8_t)s->B( 7) * (uint8_t)d->B( 7));
1374 #if SHIFT == 1
1375     d->W(4) = satsw((int8_t)s->B( 8) * (uint8_t)d->B( 8) +
1376                     (int8_t)s->B( 9) * (uint8_t)d->B( 9));
1377     d->W(5) = satsw((int8_t)s->B(10) * (uint8_t)d->B(10) +
1378                     (int8_t)s->B(11) * (uint8_t)d->B(11));
1379     d->W(6) = satsw((int8_t)s->B(12) * (uint8_t)d->B(12) +
1380                     (int8_t)s->B(13) * (uint8_t)d->B(13));
1381     d->W(7) = satsw((int8_t)s->B(14) * (uint8_t)d->B(14) +
1382                     (int8_t)s->B(15) * (uint8_t)d->B(15));
1383 #endif
1384 }
1385
1386 void glue(helper_phsubw, SUFFIX) (Reg *d, Reg *s)
1387 {
1388     d->W(0) = (int16_t)d->W(0) - (int16_t)d->W(1);
1389     d->W(1) = (int16_t)d->W(2) - (int16_t)d->W(3);
1390     XMM_ONLY(d->W(2) = (int16_t)d->W(4) - (int16_t)d->W(5));
1391     XMM_ONLY(d->W(3) = (int16_t)d->W(6) - (int16_t)d->W(7));
1392     d->W((2 << SHIFT) + 0) = (int16_t)s->W(0) - (int16_t)s->W(1);
1393     d->W((2 << SHIFT) + 1) = (int16_t)s->W(2) - (int16_t)s->W(3);
1394     XMM_ONLY(d->W(6) = (int16_t)s->W(4) - (int16_t)s->W(5));
1395     XMM_ONLY(d->W(7) = (int16_t)s->W(6) - (int16_t)s->W(7));
1396 }
1397
1398 void glue(helper_phsubd, SUFFIX) (Reg *d, Reg *s)
1399 {
1400     d->L(0) = (int32_t)d->L(0) - (int32_t)d->L(1);
1401     XMM_ONLY(d->L(1) = (int32_t)d->L(2) - (int32_t)d->L(3));
1402     d->L((1 << SHIFT) + 0) = (int32_t)s->L(0) - (int32_t)s->L(1);
1403     XMM_ONLY(d->L(3) = (int32_t)s->L(2) - (int32_t)s->L(3));
1404 }
1405
1406 void glue(helper_phsubsw, SUFFIX) (Reg *d, Reg *s)
1407 {
1408     d->W(0) = satsw((int16_t)d->W(0) - (int16_t)d->W(1));
1409     d->W(1) = satsw((int16_t)d->W(2) - (int16_t)d->W(3));
1410     XMM_ONLY(d->W(2) = satsw((int16_t)d->W(4) - (int16_t)d->W(5)));
1411     XMM_ONLY(d->W(3) = satsw((int16_t)d->W(6) - (int16_t)d->W(7)));
1412     d->W((2 << SHIFT) + 0) = satsw((int16_t)s->W(0) - (int16_t)s->W(1));
1413     d->W((2 << SHIFT) + 1) = satsw((int16_t)s->W(2) - (int16_t)s->W(3));
1414     XMM_ONLY(d->W(6) = satsw((int16_t)s->W(4) - (int16_t)s->W(5)));
1415     XMM_ONLY(d->W(7) = satsw((int16_t)s->W(6) - (int16_t)s->W(7)));
1416 }
1417
1418 #define FABSB(_, x) x > INT8_MAX  ? -(int8_t ) x : x
1419 #define FABSW(_, x) x > INT16_MAX ? -(int16_t) x : x
1420 #define FABSL(_, x) x > INT32_MAX ? -(int32_t) x : x
1421 SSE_HELPER_B(helper_pabsb, FABSB)
1422 SSE_HELPER_W(helper_pabsw, FABSW)
1423 SSE_HELPER_L(helper_pabsd, FABSL)
1424
1425 #define FMULHRSW(d, s) ((int16_t) d * (int16_t) s + 0x4000) >> 15
1426 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1427
1428 #define FSIGNB(d, s) s <= INT8_MAX  ? s ? d : 0 : -(int8_t ) d
1429 #define FSIGNW(d, s) s <= INT16_MAX ? s ? d : 0 : -(int16_t) d
1430 #define FSIGNL(d, s) s <= INT32_MAX ? s ? d : 0 : -(int32_t) d
1431 SSE_HELPER_B(helper_psignb, FSIGNB)
1432 SSE_HELPER_W(helper_psignw, FSIGNW)
1433 SSE_HELPER_L(helper_psignd, FSIGNL)
1434
1435 void glue(helper_palignr, SUFFIX) (Reg *d, Reg *s, int32_t shift)
1436 {
1437     Reg r;
1438
1439     /* XXX could be checked during translation */
1440     if (shift >= (16 << SHIFT)) {
1441         r.Q(0) = 0;
1442         XMM_ONLY(r.Q(1) = 0);
1443     } else {
1444         shift <<= 3;
1445 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1446 #if SHIFT == 0
1447         r.Q(0) = SHR(s->Q(0), shift -   0) |
1448                  SHR(d->Q(0), shift -  64);
1449 #else
1450         r.Q(0) = SHR(s->Q(0), shift -   0) |
1451                  SHR(s->Q(1), shift -  64) |
1452                  SHR(d->Q(0), shift - 128) |
1453                  SHR(d->Q(1), shift - 192);
1454         r.Q(1) = SHR(s->Q(0), shift +  64) |
1455                  SHR(s->Q(1), shift -   0) |
1456                  SHR(d->Q(0), shift -  64) |
1457                  SHR(d->Q(1), shift - 128);
1458 #endif
1459 #undef SHR
1460     }
1461
1462     *d = r;
1463 }
1464
1465 #define XMM0 env->xmm_regs[0]
1466
1467 #if SHIFT == 1
1468 #define SSE_HELPER_V(name, elem, num, F)\
1469 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1470 {\
1471     d->elem(0) = F(d->elem(0), s->elem(0), XMM0.elem(0));\
1472     d->elem(1) = F(d->elem(1), s->elem(1), XMM0.elem(1));\
1473     if (num > 2) {\
1474         d->elem(2) = F(d->elem(2), s->elem(2), XMM0.elem(2));\
1475         d->elem(3) = F(d->elem(3), s->elem(3), XMM0.elem(3));\
1476         if (num > 4) {\
1477             d->elem(4) = F(d->elem(4), s->elem(4), XMM0.elem(4));\
1478             d->elem(5) = F(d->elem(5), s->elem(5), XMM0.elem(5));\
1479             d->elem(6) = F(d->elem(6), s->elem(6), XMM0.elem(6));\
1480             d->elem(7) = F(d->elem(7), s->elem(7), XMM0.elem(7));\
1481             if (num > 8) {\
1482                 d->elem(8) = F(d->elem(8), s->elem(8), XMM0.elem(8));\
1483                 d->elem(9) = F(d->elem(9), s->elem(9), XMM0.elem(9));\
1484                 d->elem(10) = F(d->elem(10), s->elem(10), XMM0.elem(10));\
1485                 d->elem(11) = F(d->elem(11), s->elem(11), XMM0.elem(11));\
1486                 d->elem(12) = F(d->elem(12), s->elem(12), XMM0.elem(12));\
1487                 d->elem(13) = F(d->elem(13), s->elem(13), XMM0.elem(13));\
1488                 d->elem(14) = F(d->elem(14), s->elem(14), XMM0.elem(14));\
1489                 d->elem(15) = F(d->elem(15), s->elem(15), XMM0.elem(15));\
1490             }\
1491         }\
1492     }\
1493 }
1494
1495 #define SSE_HELPER_I(name, elem, num, F)\
1496 void glue(name, SUFFIX) (Reg *d, Reg *s, uint32_t imm)\
1497 {\
1498     d->elem(0) = F(d->elem(0), s->elem(0), ((imm >> 0) & 1));\
1499     d->elem(1) = F(d->elem(1), s->elem(1), ((imm >> 1) & 1));\
1500     if (num > 2) {\
1501         d->elem(2) = F(d->elem(2), s->elem(2), ((imm >> 2) & 1));\
1502         d->elem(3) = F(d->elem(3), s->elem(3), ((imm >> 3) & 1));\
1503         if (num > 4) {\
1504             d->elem(4) = F(d->elem(4), s->elem(4), ((imm >> 4) & 1));\
1505             d->elem(5) = F(d->elem(5), s->elem(5), ((imm >> 5) & 1));\
1506             d->elem(6) = F(d->elem(6), s->elem(6), ((imm >> 6) & 1));\
1507             d->elem(7) = F(d->elem(7), s->elem(7), ((imm >> 7) & 1));\
1508             if (num > 8) {\
1509                 d->elem(8) = F(d->elem(8), s->elem(8), ((imm >> 8) & 1));\
1510                 d->elem(9) = F(d->elem(9), s->elem(9), ((imm >> 9) & 1));\
1511                 d->elem(10) = F(d->elem(10), s->elem(10), ((imm >> 10) & 1));\
1512                 d->elem(11) = F(d->elem(11), s->elem(11), ((imm >> 11) & 1));\
1513                 d->elem(12) = F(d->elem(12), s->elem(12), ((imm >> 12) & 1));\
1514                 d->elem(13) = F(d->elem(13), s->elem(13), ((imm >> 13) & 1));\
1515                 d->elem(14) = F(d->elem(14), s->elem(14), ((imm >> 14) & 1));\
1516                 d->elem(15) = F(d->elem(15), s->elem(15), ((imm >> 15) & 1));\
1517             }\
1518         }\
1519     }\
1520 }
1521
1522 /* SSE4.1 op helpers */
1523 #define FBLENDVB(d, s, m) (m & 0x80) ? s : d
1524 #define FBLENDVPS(d, s, m) (m & 0x80000000) ? s : d
1525 #define FBLENDVPD(d, s, m) (m & 0x8000000000000000LL) ? s : d
1526 SSE_HELPER_V(helper_pblendvb, B, 16, FBLENDVB)
1527 SSE_HELPER_V(helper_blendvps, L, 4, FBLENDVPS)
1528 SSE_HELPER_V(helper_blendvpd, Q, 2, FBLENDVPD)
1529
1530 void glue(helper_ptest, SUFFIX) (Reg *d, Reg *s)
1531 {
1532     uint64_t zf = (s->Q(0) &  d->Q(0)) | (s->Q(1) &  d->Q(1));
1533     uint64_t cf = (s->Q(0) & ~d->Q(0)) | (s->Q(1) & ~d->Q(1));
1534
1535     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1536 }
1537
1538 #define SSE_HELPER_F(name, elem, num, F)\
1539 void glue(name, SUFFIX) (Reg *d, Reg *s)\
1540 {\
1541     d->elem(0) = F(0);\
1542     d->elem(1) = F(1);\
1543     if (num > 2) {\
1544         d->elem(2) = F(2);\
1545         d->elem(3) = F(3);\
1546         if (num > 4) {\
1547             d->elem(4) = F(4);\
1548             d->elem(5) = F(5);\
1549             d->elem(6) = F(6);\
1550             d->elem(7) = F(7);\
1551         }\
1552     }\
1553 }
1554
1555 SSE_HELPER_F(helper_pmovsxbw, W, 8, (int8_t) s->B)
1556 SSE_HELPER_F(helper_pmovsxbd, L, 4, (int8_t) s->B)
1557 SSE_HELPER_F(helper_pmovsxbq, Q, 2, (int8_t) s->B)
1558 SSE_HELPER_F(helper_pmovsxwd, L, 4, (int16_t) s->W)
1559 SSE_HELPER_F(helper_pmovsxwq, Q, 2, (int16_t) s->W)
1560 SSE_HELPER_F(helper_pmovsxdq, Q, 2, (int32_t) s->L)
1561 SSE_HELPER_F(helper_pmovzxbw, W, 8, s->B)
1562 SSE_HELPER_F(helper_pmovzxbd, L, 4, s->B)
1563 SSE_HELPER_F(helper_pmovzxbq, Q, 2, s->B)
1564 SSE_HELPER_F(helper_pmovzxwd, L, 4, s->W)
1565 SSE_HELPER_F(helper_pmovzxwq, Q, 2, s->W)
1566 SSE_HELPER_F(helper_pmovzxdq, Q, 2, s->L)
1567
1568 void glue(helper_pmuldq, SUFFIX) (Reg *d, Reg *s)
1569 {
1570     d->Q(0) = (int64_t) (int32_t) d->L(0) * (int32_t) s->L(0);
1571     d->Q(1) = (int64_t) (int32_t) d->L(2) * (int32_t) s->L(2);
1572 }
1573
1574 #define FCMPEQQ(d, s) d == s ? -1 : 0
1575 SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
1576
1577 void glue(helper_packusdw, SUFFIX) (Reg *d, Reg *s)
1578 {
1579     d->W(0) = satuw((int32_t) d->L(0));
1580     d->W(1) = satuw((int32_t) d->L(1));
1581     d->W(2) = satuw((int32_t) d->L(2));
1582     d->W(3) = satuw((int32_t) d->L(3));
1583     d->W(4) = satuw((int32_t) s->L(0));
1584     d->W(5) = satuw((int32_t) s->L(1));
1585     d->W(6) = satuw((int32_t) s->L(2));
1586     d->W(7) = satuw((int32_t) s->L(3));
1587 }
1588
1589 #define FMINSB(d, s) MIN((int8_t) d, (int8_t) s)
1590 #define FMINSD(d, s) MIN((int32_t) d, (int32_t) s)
1591 #define FMAXSB(d, s) MAX((int8_t) d, (int8_t) s)
1592 #define FMAXSD(d, s) MAX((int32_t) d, (int32_t) s)
1593 SSE_HELPER_B(helper_pminsb, FMINSB)
1594 SSE_HELPER_L(helper_pminsd, FMINSD)
1595 SSE_HELPER_W(helper_pminuw, MIN)
1596 SSE_HELPER_L(helper_pminud, MIN)
1597 SSE_HELPER_B(helper_pmaxsb, FMAXSB)
1598 SSE_HELPER_L(helper_pmaxsd, FMAXSD)
1599 SSE_HELPER_W(helper_pmaxuw, MAX)
1600 SSE_HELPER_L(helper_pmaxud, MAX)
1601
1602 #define FMULLD(d, s) (int32_t) d * (int32_t) s
1603 SSE_HELPER_L(helper_pmulld, FMULLD)
1604
1605 void glue(helper_phminposuw, SUFFIX) (Reg *d, Reg *s)
1606 {
1607     int idx = 0;
1608
1609     if (s->W(1) < s->W(idx))
1610         idx = 1;
1611     if (s->W(2) < s->W(idx))
1612         idx = 2;
1613     if (s->W(3) < s->W(idx))
1614         idx = 3;
1615     if (s->W(4) < s->W(idx))
1616         idx = 4;
1617     if (s->W(5) < s->W(idx))
1618         idx = 5;
1619     if (s->W(6) < s->W(idx))
1620         idx = 6;
1621     if (s->W(7) < s->W(idx))
1622         idx = 7;
1623
1624     d->Q(1) = 0;
1625     d->L(1) = 0;
1626     d->W(1) = idx;
1627     d->W(0) = s->W(idx);
1628 }
1629
1630 void glue(helper_roundps, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1631 {
1632     signed char prev_rounding_mode;
1633
1634     prev_rounding_mode = env->sse_status.float_rounding_mode;
1635     if (!(mode & (1 << 2)))
1636         switch (mode & 3) {
1637         case 0:
1638             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1639             break;
1640         case 1:
1641             set_float_rounding_mode(float_round_down, &env->sse_status);
1642             break;
1643         case 2:
1644             set_float_rounding_mode(float_round_up, &env->sse_status);
1645             break;
1646         case 3:
1647             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1648             break;
1649         }
1650
1651     d->L(0) = float64_round_to_int(s->L(0), &env->sse_status);
1652     d->L(1) = float64_round_to_int(s->L(1), &env->sse_status);
1653     d->L(2) = float64_round_to_int(s->L(2), &env->sse_status);
1654     d->L(3) = float64_round_to_int(s->L(3), &env->sse_status);
1655
1656 #if 0 /* TODO */
1657     if (mode & (1 << 3))
1658         set_float_exception_flags(
1659                         get_float_exception_flags(&env->sse_status) &
1660                         ~float_flag_inexact,
1661                         &env->sse_status);
1662 #endif
1663     env->sse_status.float_rounding_mode = prev_rounding_mode;
1664 }
1665
1666 void glue(helper_roundpd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1667 {
1668     signed char prev_rounding_mode;
1669
1670     prev_rounding_mode = env->sse_status.float_rounding_mode;
1671     if (!(mode & (1 << 2)))
1672         switch (mode & 3) {
1673         case 0:
1674             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1675             break;
1676         case 1:
1677             set_float_rounding_mode(float_round_down, &env->sse_status);
1678             break;
1679         case 2:
1680             set_float_rounding_mode(float_round_up, &env->sse_status);
1681             break;
1682         case 3:
1683             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1684             break;
1685         }
1686
1687     d->Q(0) = float64_round_to_int(s->Q(0), &env->sse_status);
1688     d->Q(1) = float64_round_to_int(s->Q(1), &env->sse_status);
1689
1690 #if 0 /* TODO */
1691     if (mode & (1 << 3))
1692         set_float_exception_flags(
1693                         get_float_exception_flags(&env->sse_status) &
1694                         ~float_flag_inexact,
1695                         &env->sse_status);
1696 #endif
1697     env->sse_status.float_rounding_mode = prev_rounding_mode;
1698 }
1699
1700 void glue(helper_roundss, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1701 {
1702     signed char prev_rounding_mode;
1703
1704     prev_rounding_mode = env->sse_status.float_rounding_mode;
1705     if (!(mode & (1 << 2)))
1706         switch (mode & 3) {
1707         case 0:
1708             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1709             break;
1710         case 1:
1711             set_float_rounding_mode(float_round_down, &env->sse_status);
1712             break;
1713         case 2:
1714             set_float_rounding_mode(float_round_up, &env->sse_status);
1715             break;
1716         case 3:
1717             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1718             break;
1719         }
1720
1721     d->L(0) = float64_round_to_int(s->L(0), &env->sse_status);
1722
1723 #if 0 /* TODO */
1724     if (mode & (1 << 3))
1725         set_float_exception_flags(
1726                         get_float_exception_flags(&env->sse_status) &
1727                         ~float_flag_inexact,
1728                         &env->sse_status);
1729 #endif
1730     env->sse_status.float_rounding_mode = prev_rounding_mode;
1731 }
1732
1733 void glue(helper_roundsd, SUFFIX) (Reg *d, Reg *s, uint32_t mode)
1734 {
1735     signed char prev_rounding_mode;
1736
1737     prev_rounding_mode = env->sse_status.float_rounding_mode;
1738     if (!(mode & (1 << 2)))
1739         switch (mode & 3) {
1740         case 0:
1741             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1742             break;
1743         case 1:
1744             set_float_rounding_mode(float_round_down, &env->sse_status);
1745             break;
1746         case 2:
1747             set_float_rounding_mode(float_round_up, &env->sse_status);
1748             break;
1749         case 3:
1750             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1751             break;
1752         }
1753
1754     d->Q(0) = float64_round_to_int(s->Q(0), &env->sse_status);
1755
1756 #if 0 /* TODO */
1757     if (mode & (1 << 3))
1758         set_float_exception_flags(
1759                         get_float_exception_flags(&env->sse_status) &
1760                         ~float_flag_inexact,
1761                         &env->sse_status);
1762 #endif
1763     env->sse_status.float_rounding_mode = prev_rounding_mode;
1764 }
1765
1766 #define FBLENDP(d, s, m) m ? s : d
1767 SSE_HELPER_I(helper_blendps, L, 4, FBLENDP)
1768 SSE_HELPER_I(helper_blendpd, Q, 2, FBLENDP)
1769 SSE_HELPER_I(helper_pblendw, W, 8, FBLENDP)
1770
1771 void glue(helper_dpps, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1772 {
1773     float32 iresult = 0 /*float32_zero*/;
1774
1775     if (mask & (1 << 4))
1776         iresult = float32_add(iresult,
1777                         float32_mul(d->L(0), s->L(0), &env->sse_status),
1778                         &env->sse_status);
1779     if (mask & (1 << 5))
1780         iresult = float32_add(iresult,
1781                         float32_mul(d->L(1), s->L(1), &env->sse_status),
1782                         &env->sse_status);
1783     if (mask & (1 << 6))
1784         iresult = float32_add(iresult,
1785                         float32_mul(d->L(2), s->L(2), &env->sse_status),
1786                         &env->sse_status);
1787     if (mask & (1 << 7))
1788         iresult = float32_add(iresult,
1789                         float32_mul(d->L(3), s->L(3), &env->sse_status),
1790                         &env->sse_status);
1791     d->L(0) = (mask & (1 << 0)) ? iresult : 0 /*float32_zero*/;
1792     d->L(1) = (mask & (1 << 1)) ? iresult : 0 /*float32_zero*/;
1793     d->L(2) = (mask & (1 << 2)) ? iresult : 0 /*float32_zero*/;
1794     d->L(3) = (mask & (1 << 3)) ? iresult : 0 /*float32_zero*/;
1795 }
1796
1797 void glue(helper_dppd, SUFFIX) (Reg *d, Reg *s, uint32_t mask)
1798 {
1799     float64 iresult = 0 /*float64_zero*/;
1800
1801     if (mask & (1 << 4))
1802         iresult = float64_add(iresult,
1803                         float64_mul(d->Q(0), s->Q(0), &env->sse_status),
1804                         &env->sse_status);
1805     if (mask & (1 << 5))
1806         iresult = float64_add(iresult,
1807                         float64_mul(d->Q(1), s->Q(1), &env->sse_status),
1808                         &env->sse_status);
1809     d->Q(0) = (mask & (1 << 0)) ? iresult : 0 /*float64_zero*/;
1810     d->Q(1) = (mask & (1 << 1)) ? iresult : 0 /*float64_zero*/;
1811 }
1812
1813 void glue(helper_mpsadbw, SUFFIX) (Reg *d, Reg *s, uint32_t offset)
1814 {
1815     int s0 = (offset & 3) << 2;
1816     int d0 = (offset & 4) << 0;
1817     int i;
1818     Reg r;
1819
1820     for (i = 0; i < 8; i++, d0++) {
1821         r.W(i) = 0;
1822         r.W(i) += abs1(d->B(d0 + 0) - s->B(s0 + 0));
1823         r.W(i) += abs1(d->B(d0 + 1) - s->B(s0 + 1));
1824         r.W(i) += abs1(d->B(d0 + 2) - s->B(s0 + 2));
1825         r.W(i) += abs1(d->B(d0 + 3) - s->B(s0 + 3));
1826     }
1827
1828     *d = r;
1829 }
1830
1831 /* SSE4.2 op helpers */
1832 /* it's unclear whether signed or unsigned */
1833 #define FCMPGTQ(d, s) d > s ? -1 : 0
1834 SSE_HELPER_Q(helper_pcmpgtq, FCMPGTQ)
1835
1836 static inline int pcmp_elen(int reg, uint32_t ctrl)
1837 {
1838     int val;
1839
1840     /* Presence of REX.W is indicated by a bit higher than 7 set */
1841     if (ctrl >> 8)
1842         val = abs1((int64_t) env->regs[reg]);
1843     else
1844         val = abs1((int32_t) env->regs[reg]);
1845
1846     if (ctrl & 1) {
1847         if (val > 8)
1848             return 8;
1849     } else
1850         if (val > 16)
1851             return 16;
1852
1853     return val;
1854 }
1855
1856 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1857 {
1858     int val = 0;
1859
1860     if (ctrl & 1) {
1861         while (val < 8 && r->W(val))
1862             val++;
1863     } else
1864         while (val < 16 && r->B(val))
1865             val++;
1866
1867     return val;
1868 }
1869
1870 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1871 {
1872     switch ((ctrl >> 0) & 3) {
1873     case 0:
1874         return r->B(i);
1875     case 1:
1876         return r->W(i);
1877     case 2:
1878         return (int8_t) r->B(i);
1879     case 3:
1880     default:
1881         return (int16_t) r->W(i);
1882     }
1883 }
1884
1885 static inline unsigned pcmpxstrx(Reg *d, Reg *s,
1886                 int8_t ctrl, int valids, int validd)
1887 {
1888     unsigned int res = 0;
1889     int v;
1890     int j, i;
1891     int upper = (ctrl & 1) ? 7 : 15;
1892
1893     valids--;
1894     validd--;
1895
1896     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1897
1898     switch ((ctrl >> 2) & 3) {
1899     case 0:
1900         for (j = valids; j >= 0; j--) {
1901             res <<= 1;
1902             v = pcmp_val(s, ctrl, j);
1903             for (i = validd; i >= 0; i--)
1904                 res |= (v == pcmp_val(d, ctrl, i));
1905         }
1906         break;
1907     case 1:
1908         for (j = valids; j >= 0; j--) {
1909             res <<= 1;
1910             v = pcmp_val(s, ctrl, j);
1911             for (i = ((validd - 1) | 1); i >= 0; i -= 2)
1912                 res |= (pcmp_val(d, ctrl, i - 0) <= v &&
1913                         pcmp_val(d, ctrl, i - 1) >= v);
1914         }
1915         break;
1916     case 2:
1917         res = (2 << (upper - MAX(valids, validd))) - 1;
1918         res <<= MAX(valids, validd) - MIN(valids, validd);
1919         for (i = MIN(valids, validd); i >= 0; i--) {
1920             res <<= 1;
1921             v = pcmp_val(s, ctrl, i);
1922             res |= (v == pcmp_val(d, ctrl, i));
1923         }
1924         break;
1925     case 3:
1926         for (j = valids - validd; j >= 0; j--) {
1927             res <<= 1;
1928             res |= 1;
1929             for (i = MIN(upper - j, validd); i >= 0; i--)
1930                 res &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
1931         }
1932         break;
1933     }
1934
1935     switch ((ctrl >> 4) & 3) {
1936     case 1:
1937         res ^= (2 << upper) - 1;
1938         break;
1939     case 3:
1940         res ^= (2 << valids) - 1;
1941         break;
1942     }
1943
1944     if (res)
1945        CC_SRC |= CC_C;
1946     if (res & 1)
1947        CC_SRC |= CC_O;
1948
1949     return res;
1950 }
1951
1952 static inline int rffs1(unsigned int val)
1953 {
1954     int ret = 1, hi;
1955
1956     for (hi = sizeof(val) * 4; hi; hi /= 2)
1957         if (val >> hi) {
1958             val >>= hi;
1959             ret += hi;
1960         }
1961
1962     return ret;
1963 }
1964
1965 static inline int ffs1(unsigned int val)
1966 {
1967     int ret = 1, hi;
1968
1969     for (hi = sizeof(val) * 4; hi; hi /= 2)
1970         if (val << hi) {
1971             val <<= hi;
1972             ret += hi;
1973         }
1974
1975     return ret;
1976 }
1977
1978 void glue(helper_pcmpestri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1979 {
1980     unsigned int res = pcmpxstrx(d, s, ctrl,
1981                     pcmp_elen(R_EDX, ctrl),
1982                     pcmp_elen(R_EAX, ctrl));
1983
1984     if (res)
1985         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
1986     else
1987         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
1988 }
1989
1990 void glue(helper_pcmpestrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
1991 {
1992     int i;
1993     unsigned int res = pcmpxstrx(d, s, ctrl,
1994                     pcmp_elen(R_EDX, ctrl),
1995                     pcmp_elen(R_EAX, ctrl));
1996
1997     if ((ctrl >> 6) & 1) {
1998         if (ctrl & 1)
1999             for (i = 0; i <= 8; i--, res >>= 1)
2000                 d->W(i) = (res & 1) ? ~0 : 0;
2001         else
2002             for (i = 0; i <= 16; i--, res >>= 1)
2003                 d->B(i) = (res & 1) ? ~0 : 0;
2004     } else {
2005         d->Q(1) = 0;
2006         d->Q(0) = res;
2007     }
2008 }
2009
2010 void glue(helper_pcmpistri, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
2011 {
2012     unsigned int res = pcmpxstrx(d, s, ctrl,
2013                     pcmp_ilen(s, ctrl),
2014                     pcmp_ilen(d, ctrl));
2015
2016     if (res)
2017         env->regs[R_ECX] = ((ctrl & (1 << 6)) ? rffs1 : ffs1)(res) - 1;
2018     else
2019         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2020 }
2021
2022 void glue(helper_pcmpistrm, SUFFIX) (Reg *d, Reg *s, uint32_t ctrl)
2023 {
2024     int i;
2025     unsigned int res = pcmpxstrx(d, s, ctrl,
2026                     pcmp_ilen(s, ctrl),
2027                     pcmp_ilen(d, ctrl));
2028
2029     if ((ctrl >> 6) & 1) {
2030         if (ctrl & 1)
2031             for (i = 0; i <= 8; i--, res >>= 1)
2032                 d->W(i) = (res & 1) ? ~0 : 0;
2033         else
2034             for (i = 0; i <= 16; i--, res >>= 1)
2035                 d->B(i) = (res & 1) ? ~0 : 0;
2036     } else {
2037         d->Q(1) = 0;
2038         d->Q(0) = res;
2039     }
2040 }
2041
2042 #define CRCPOLY        0x1edc6f41
2043 #define CRCPOLY_BITREV 0x82f63b78
2044 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2045 {
2046     target_ulong crc = (msg & ((target_ulong) -1 >>
2047                             (TARGET_LONG_BITS - len))) ^ crc1;
2048
2049     while (len--)
2050         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2051
2052     return crc;
2053 }
2054
2055 #define POPMASK(i)     ((target_ulong) -1 / ((1LL << (1 << i)) + 1))
2056 #define POPCOUNT(n, i) (n & POPMASK(i)) + ((n >> (1 << i)) & POPMASK(i))
2057 target_ulong helper_popcnt(target_ulong n, uint32_t type)
2058 {
2059     CC_SRC = n ? 0 : CC_Z;
2060
2061     n = POPCOUNT(n, 0);
2062     n = POPCOUNT(n, 1);
2063     n = POPCOUNT(n, 2);
2064     n = POPCOUNT(n, 3);
2065     if (type == 1)
2066         return n & 0xff;
2067
2068     n = POPCOUNT(n, 4);
2069 #ifndef TARGET_X86_64
2070     return n;
2071 #else
2072     if (type == 2)
2073         return n & 0xff;
2074
2075     return POPCOUNT(n, 5);
2076 #endif
2077 }
2078 #endif
2079
2080 #undef SHIFT
2081 #undef XMM_ONLY
2082 #undef Reg
2083 #undef B
2084 #undef W
2085 #undef L
2086 #undef Q
2087 #undef SUFFIX